Realtime AI News
Alibaba unveils Wan3.0 with 30-second video generation and multimodal inputs
Alibaba has unveiled Wan3.0, its video generation model that supports up to 30 seconds of output and multimodal inputs, according to Pokde.Net. The release extends Alibaba's push into long-form, controllable video generation and intensifies competition in the field.
Alibaba has unveiled Wan3.0, its video generation model that supports up to 30 seconds of video generation and multimodal inputs, according to Pokde.Net.
The 30-second generation length is a clear step up from typical short video clips, meaning the model can cover more complete narrative segments rather than isolated shots.
Multimodal inputs let users guide video generation through images, text and other modalities at the same time, giving creators more freedom and consistency in their output.
The release is another major upgrade in Alibaba's video generation push. The Wan series has already built a user base in the open-source community, and the new version extends the company's broader strategy in multimodal large models.
With competition in video generation intensifying, long-form generation and multimodal control are exactly the fronts where models are being compared, and Wan3.0 raises the stakes in the field.
What to watch next: whether Wan3.0 is open-sourced, how it performs on benchmarks, and how it lands in applications such as e-commerce and marketing.
Why it matters
Wan3.0 pushes video generation to 30 seconds with multimodal control, potentially accelerating adoption in e-commerce, marketing and other content-heavy applications.
Nearby Updates
All08/12, 00:47
Databricks acquires Electric to give every AI agent its own Postgres database
Databricks has acquired Electric, a database startup, with the goal of giving every AI agent its own Postgres database, according to The New Stack. The move embeds database infrastructure directly into AI agent tooling, lowering the barrier for enterprises deploying agents at scale.
08/12, 00:48
Report: Nvidia building 1-trillion-parameter Nemotron 4 to rival open AI models
Nvidia is building a 1-trillion-parameter Nemotron 4 model to rival open AI models, according to The Information. If confirmed, it would be a major upgrade to Nvidia's open-source model line and a sign that the chip giant is investing more heavily in the model layer.
08/12, 01:00
Google's AMIE medical AI demonstrates real-time clinical video consultations in first-of-its-kind study
Google has shown that AMIE, its research medical AI system, can conduct real-time clinical video consultations in a first-of-its-kind study. In simulated clinical settings, the system engages patients through live video, marking a step beyond text-based medical AI toward multimodal clinical interaction.
08/12, 00:25
An unreleased Anthropic model made surprising progress on the Riemann hypothesis, TechCrunch reports
TechCrunch reports that an unreleased Anthropic model made more progress than expected on the Riemann hypothesis, one of mathematics' biggest unsolved problems, though Anthropic has not solved it. The report signals that frontier mathematical reasoning is still improving quickly, and raises questions about when the model will ship.